The email the assistant is writing now sits beside the chat - and it won’t overwrite your words

Asking the assistant to draft an email used to give you the email in the conversation. Ask for it shorter and you got a second one. Ask again and you had three, all nearly identical, scrolling away above you, and no way to tell which one was real. There is now one draft, in a panel beside the chat. It stays put while the conversation scrolls, it survives a reload, and every change goes into that one email rather than producing another copy. The part that matters is what happens when you’ve typed into it. Once you’ve touched the draft, the assistant stops writing into it. Ask for a rewrite and the change arrives as a suggestion - your version stays exactly as you left it, with the proposed wording shown above it, struck-through before and plain after, and two buttons: use theirs, or keep yours. Nothing is lost either way. That is deliberate and it isn’t caution: an assistant that silently replaces a half-written email in a panel you weren’t looking at is one you switch off, and rightly. It also won’t merge the two versions into sentences neither of you wrote, which looks like it worked and is worse. It says so honestly, too - “I’ve suggested a rewrite, it’s waiting for you” rather than “done”. Send goes from the panel, and sends exactly what you were reading: if anything moved the draft between you reading it and pressing the button, it tells you and asks you to look again rather than posting something you never saw. It names what’s missing - no subject, no recipient, empty body - while you’re writing, instead of at the moment you press Send. Your permission to send from that address, and whether the recipient has unsubscribed, are re-checked at the moment of sending, on exactly the same path the rest of the product uses.